Saint-Maurice et al. BMC Public Health 2014, 14:103 http://www.biomedcentral.com/1471-2458/14/103
RESEARCH ARTICLE
Open Access
Moderating influences of baseline activity levels in school physical activity programming for children: the Ready for Recess project Pedro F Saint-Maurice1*, Gregory J Welk1, Daniel W Russell2 and Jennifer Huberty3
Abstract Background: A limitation of traditional outcome studies from behavioral interventions is the lack of attention given to evaluating the influence of moderating variables. This study examined possible moderation effect of baseline activity levels on physical activity change as a result of the Ready for Recess intervention. Methods: Ready for Recess (August 2009-September 2010) was a controlled trial with twelve schools randomly assigned to one of four conditions: control group, staff supervision, equipment availability, and the combination of staff supervision and equipment availability. A total of 393 children (181 boys and 212 girls) from grades 3 through 6 (8–11 years old) were asked to wear an Actigraph monitor during school time on 4–5 days of the week. Assessments were conducted at baseline (before intervention) and post intervention (after intervention). Results: Initial MVPA moderated the effect of Staff supervision (β = −0.47%; p < .05), but not Equipment alone and Staff + Equipment (p > .05). Participants in the Staff condition that were 1 standard deviation (SD) below the mean for baseline MVPA (classified as “low active”) had lower MVPA levels at post-intervention when compared with their low active peers in the control condition (Mean diff = −10.8 ± 2.9%; p = .005). High active individuals (+1SD above the mean) in the Equipment treatment also had lower MVPA values at post-intervention when compared with their highly active peers in the control group (Mean diff = −9.5 ± 2.9%; p = .009). Conclusions: These results indicate that changes in MVPA levels at post-intervention were reduced in highly active participants when recess staff supervision was provided. In this study, initial MVPA moderated the effect of Staff supervision on children’s MVPA after 6 months of intervention. Staff training should include how to work with inactive youth but also how to assure that active children remain active. Keywords: PA promotion, MVPA, Recess, Youth
Background There is considerable public health interest in developing and testing strategies to help youth be more physically active [1]. School-based interventions are a common target due to the opportunity to reach large numbers of youth and the availability of staff and resources [2], but unfortunately, the success of school-based programming is limited [1,3]. A limitation of traditional analyses of school-based interventions is that the focus is on evaluating only the overall group-level effects. This provides a limited perspective * Correspondence: [email protected] 1 Iowa State University, Department of Kinesiology, Ames, IA 50011, USA Full list of author information is available at the end of the article
of the outcomes since it is possible for interventions to have differential effects within the sample population. An intervention, for example, might have benefits for inactive youth but not for others. It is therefore important to conduct follow-up evaluations of the data collected to examine possible moderating influences. Members of our team recently reported on the outcomes from a recess-based intervention, the Readyfor-Recess (R4R) project, a controlled trial evaluating environmental modifications to promote physical activity during recess [4]. Recess plays a critical role in children’s development and well being [5,6] and recess time has also been shown to make important contributions to children’s school and total daily activity [7-10].
© 2014 Saint-Maurice et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.
Saint-Maurice et al. BMC Public Health 2014, 14:103 http://www.biomedcentral.com/1471-2458/14/103
Ridgers et al., 2012, recently conducted a systematic review to examine activity levels during recess. The review emphasized the need to explore different activity promotion strategies and how these can impact the behaviors of important subgroups (e.g. boys vs. girls) [11]. Other recess-based studies have examined the benefits of adult supervision [12], and equipment availability during recess [13], but a unique aspect of R4R was that it evaluated the independent and combined effects of adult supervision and equipment on physical activity during recess. Schools in the R4R study (n = 12) were randomly assigned to one of four conditions: control group, staff supervision, provision of equipment, and the combination of staff supervision and equipment. The combination of staff supervision and equipment availability led to a significant increase of moderate-to-vigorous physical activity (MVPA) in boys (+14.1%) [4]. These findings suggest that recess interventions can be successful when promoting active behaviors among groups of children. As with most clinical trials [14], the success of R4R was examined using group-level comparisons with the rate of change across time being defined as a fixed effect. This approach assumes that the amount of change across individuals is the same. While this fixed effects approach is appropriate for the overall evaluation, further work is required so that the assumption of homogeneity across individuals can be examined and to avoid what is defined as the “ecological fallacy” [15,16]. The inter-individual variability in the context of a PA intervention can be examined by allowing the impact of the interventions to be defined as random in the regression model. A test of homogeneity is used to determine if the rate of change across time is consistent across individuals. If there is variability between individuals one can try to explain this variability and thereby better understand individual determinants of PA. This approach can be undertaken using latent growth curve analysis, which is similar to the classic fixed effects model but with the advantage that researchers can specify predictors in the model as random factors [16]. This approach makes it possible to systematically evaluate factors influencing the consistency of change across individuals. The purpose of this study was to examine if baseline levels of moderate-to-vigorous physical activity (MVPA) moderated the effects from the various R4R treatments.
Methods Intervention design
The R4R (August 2009-September 2010) project evaluated the independent and combined effects of staff training and equipment on children’s participation in MVPA during daily recess. Twelve schools were randomly assigned to one of four conditions: Staff Training (ST), Equipment (EQ), Staff Training + Equipment (STEQ) or
Page 2 of 8
a control condition. We have published a main outcome paper that provides additional details about the R4R intervention [4]. Participants
Elementary children (3rd-6th grades, 8–11 years of age) from 12 participating schools were recruited to participate in the evaluation of the R4R intervention. Demographic information obtained from the school included: age, gender, birth date, race, and free and reduced lunch status. The Ready for Recess intervention study was approved by the Institutional Review Board at University of Nebraska at Omaha. Children that returned signed consent forms were enrolled in the study. Anthropometric measures
Height and weight data on participants were obtained by the research team using standard field based techniques. However, we had to minimize the disruption of the school schedule at some of the schools and therefore were not able collect this information on the entire sample involved in the study. Physical activity - accelerometer measure
Physical Activity was assessed with the GT1M Actigraph Accelerometers (Actigraph, Pensacola, FL). Students were asked to wear the activity monitors during the school day (approximately 8:00 AM – 3:00 PM), for 4–5 consecutive days. Members of the research team put the monitors on the children each morning and took them off at the end of the school day to ensure compliance. Teachers provided logs of student attendance and a daily schedule to record time periods for recess, lunch, and physical education. The data from the Actigraphs were recorded in 5 second epochs and processed using Freedson et al. (2005) age-specific cutpoints (METs = 2.757 + (0.0015 · counts · min-1) – (0.08957 · age [yr]) – (0.000038 · counts · min1 · age [yr]) [17]. Each epoch during recess was classified as light (1.5 - 3.9 METs), moderate (MPA; 4.0 - 6.0 METs) or vigorous (VPA; > 6.0 METs) and the data were then aggregated to determine time spent in LPA, MPA, VPA and MVPA (MPA + VPA). Participants were included in the final analyses if they were at school (and had data) on at least 3 of the 5 days. Since the activity monitors were only used during school time it was assumed that students wore the monitors during all time unless indicated otherwise by school teachers. Activity outcomes were computed as an average of the valid data extracted from recess periods. The major computed outcome measure was the percent time allocated to MVPA in order to standardize for different lengths of recess.
Saint-Maurice et al. BMC Public Health 2014, 14:103 http://www.biomedcentral.com/1471-2458/14/103
Data analysis
Latent Growth Curve (LGC) models were used to account for the longitudinal nature of the data (repeatedmeasures) and potential individual differences (nested within schools) when exposed to the different treatments. The variable Trial (coded as 0 = trial 1, 1 = trial 2) was defined as a level-1 predictor (growth factor), characteristics of the children as level-2 predictors, and schools as level-3 predictors. Post-intervention average percent time in MVPA during recess (activity obtained from valid recess segments) was defined as the main outcome of interest (dependent variable). Level-2 covariates included grade (coded with 0 = 3rd grade, 1 = 4th grade, 2 = 5th grade, and 3 = 6th grade), gender (coded 0 = male and 1 = female), average recess duration (in minutes), and baseline (trial 1) percent time in MVPA during recess. The three treatments were used as level-3 predictors and coded using dummy coding (variable Staff coded 0 = No Staff, 1 = Staff; variable Equipment coded 0 = No equipment, 1 = Equipment; and variable Staff + Equipment coded 0 = No Staff + Equipment, and 1 = Staff + Equipment). Other demographic characteristics of the children such as BMI, race, or social economic status were not included since this information was not available for the entire sample. Only individuals with complete activity data at each assessment (Trial) were included in the analysis. To examine the direction, magnitude and consistency of the moderation of baseline MVPA on each intervention, least-square means comparing each treatment to the control group were computed (using the final LGC model) at three fixed values based on baseline MVPA scores: Mean - 1 standard deviation, Mean, and Mean + 1 standard deviation (see magnitude and direction of moderation effect). These three categories were operationalized as Low Active, Moderate Active, and High Active, respectively. These comparisons were performed to examine the overall trend over time (across treatments and control group) and also separately for each activity group. A total of 12 least-square means tests with their respective mean difference (Mean diff ), standard errors and p-values are reported. Results from the nested models (test of model fit for level-1 and level-2 predictors) and moderation effects were computed with 95% confidence intervals (p < .05). To account for the increased type I error associated with multiple statistical tests when comparing subgroups of activity, statistical analysis associated with least-square means were performed with a 99% confidence interval and therefore defined as significant if p-value < .01. In addition, mean differences for these analyses were considered borderline significant if p-value was between .01 and .05. Statistical procedures as described above were performed using SAS v9.2 (Cary, North Carolina).
Page 3 of 8
Results From the total sample of 667 participants, there were 393 participants (181 boys and 212 girls) who had three days of activity monitor data on both trials. The sample size differs from the main outcome results previously reported by Huberty et al., 2013, since it was important to have robust estimates of individual variability patterns across time. The distribution of participants across gender, grade and intervention is provided in Table 1. Inter-individual variability and test of covariates
Preliminary analyses were conducted to test assumptions needed for the use of LGC. We examined the variability in the MVPA scores across time to ensure that there was sufficient variability in the data. The intraclass correlation for individuals was equal to .08, while for schools it was equal to .07 and the covariance parameters indicated that intercept scores for time spent in MVPA were significantly different between individuals (see Figure 1) and schools (see Figure 2) (p < .05). Additional testing with level-1 and level-2 predictors revealed significant improvements in the fit of the model (indicated by a significant (p < .05) change in the (LLR) value). The results supported the clustered nature of the data and differences on baseline activity levels between individuals and schools. These findings support the planned model used to examine the moderator effect of baseline activity on change in MVPA levels associated with each condition. The latent growth curve model
Interpretation of the latent growth curve model b-weights (Table 2) revealed that change in MVPA from trial 1 to trial 2 in the control group (holding all the other predictors in the model constant) could be explained by initial MVPA scores. Per each unit increase in baseline MVPA score, there was a reduction of 0.76% on MVPA scores change from trial 1 to trial 2 (e.g., children in the control group with lower baseline MVPA scores had greater positive changes in MVPA from trial 1 to trial 2 while the opposite was true for children with higher baseline MVPA scores). Similar trial X baseline MVPA interactions with each condition revealed that this trend was attenuated by Table 1 Distribution (N) of gender and grade per intervention group Gender Boys
Girls
Grade 3rd
4th
5th
6th
Total
Control
47
45
20
27
24
21
92
ST
39
69
32
19
34
23
108
EQ
59
51
23
27
38
22
110
STEQ
36
47
8
22
23
30
83
Total
181
212
83
95
119
96
393
ST = Staff treatment; EQ = Equipment treatment; STEQ = Staff + Equipment treatment.
Saint-Maurice et al. BMC Public Health 2014, 14:103 http://www.biomedcentral.com/1471-2458/14/103
Page 4 of 8
Figure 1 Percent time spent in MVPA at recess for a (randomly selected) subgroup of 25 individuals. Each line represents an individual. This figure illustrates a high degree of variability in individual’s baseline scores (intercept) and change over time (slopes). This justifies the need for a random intercept term at level-2 (individual).
ST (β = 0.29 ± 0.12, p = .02) but not EQ (p = .35) or STEQ (p = .74). In other words, youth in the Staff condition that were less active at baseline had greater MVPA differences (from their respective control group) in MVPA at trial 2 when compared to their more active peers. Per each unit increase in percent MVPA at baseline, there was an attenuation of the ST effect by 0.47% in MVPA levels at trial 2. Magnitude and direction of the moderation effect
The overall mean score (across treatments and control group) for percent of time spent in MVPA at baseline was 27.5%. Predicted values were estimated by fixing baseline activity values in the final model as 27.5%, 12.5% (−1SD), and 42.4% (+1SD). These predicted activity scores represented Moderate, Low, and High Active groups, respectively. There were significant changes over time for the three groups but the patterns were very
different. The Low Active group had a significant increase in MVPA from trial 1 to trial 2 (Mean diff = 6.3 ± 0.9%; t (11) = 6.7; p < .001), while the Moderate (Mean diff = −4.5 ± 0.7%; t (11) = −6.8; p < .001) and High active group (Mean diff = −15.3 ± 0.9%; t (11) = −16.2; p < .001) decreased their MVPA levels (p < .001). Further evaluation, however, shows that the response varies depending on the treatment group. When stratified by treatment group, the mean baseline MVPA values (Medium) for the ST, EQ and STEQ treatments were 27.7%, 26.1%, and 27.1%, respectively. The corresponding values for Low Active (−1SD) were equal to, 12.2%, 11.0%, and 13.3% while values for High Active (+1SD) were 43.2%, 41.3%, and 40.8%. The values for Low Active were all similar at baseline (Trial 1) but the levels of MVPA at trial 2 in the ST schools were significantly lower than the respective Low Active control group
Figure 2 Percent time spent in MVPA at recess for the 12 schools. Each line represents a school. Similarly to Figure 1, there was a significant difference in baseline scores for school clusters, and therefore, justifying the need for random intercepts at this level.
Saint-Maurice et al. BMC Public Health 2014, 14:103 http://www.biomedcentral.com/1471-2458/14/103
Page 5 of 8
Table 2 Regression coefficients for the final latent growth curve model Intercept
Estimate
SE
−6.73
5.62 8
df
t-value p-value −1.20
0.27
b-weights Trial
20.35
3.00 759 6.78
RESEARCH ARTICLE
Open Access
Moderating influences of baseline activity levels in school physical activity programming for children: the Ready for Recess project Pedro F Saint-Maurice1*, Gregory J Welk1, Daniel W Russell2 and Jennifer Huberty3
Abstract Background: A limitation of traditional outcome studies from behavioral interventions is the lack of attention given to evaluating the influence of moderating variables. This study examined possible moderation effect of baseline activity levels on physical activity change as a result of the Ready for Recess intervention. Methods: Ready for Recess (August 2009-September 2010) was a controlled trial with twelve schools randomly assigned to one of four conditions: control group, staff supervision, equipment availability, and the combination of staff supervision and equipment availability. A total of 393 children (181 boys and 212 girls) from grades 3 through 6 (8–11 years old) were asked to wear an Actigraph monitor during school time on 4–5 days of the week. Assessments were conducted at baseline (before intervention) and post intervention (after intervention). Results: Initial MVPA moderated the effect of Staff supervision (β = −0.47%; p < .05), but not Equipment alone and Staff + Equipment (p > .05). Participants in the Staff condition that were 1 standard deviation (SD) below the mean for baseline MVPA (classified as “low active”) had lower MVPA levels at post-intervention when compared with their low active peers in the control condition (Mean diff = −10.8 ± 2.9%; p = .005). High active individuals (+1SD above the mean) in the Equipment treatment also had lower MVPA values at post-intervention when compared with their highly active peers in the control group (Mean diff = −9.5 ± 2.9%; p = .009). Conclusions: These results indicate that changes in MVPA levels at post-intervention were reduced in highly active participants when recess staff supervision was provided. In this study, initial MVPA moderated the effect of Staff supervision on children’s MVPA after 6 months of intervention. Staff training should include how to work with inactive youth but also how to assure that active children remain active. Keywords: PA promotion, MVPA, Recess, Youth
Background There is considerable public health interest in developing and testing strategies to help youth be more physically active [1]. School-based interventions are a common target due to the opportunity to reach large numbers of youth and the availability of staff and resources [2], but unfortunately, the success of school-based programming is limited [1,3]. A limitation of traditional analyses of school-based interventions is that the focus is on evaluating only the overall group-level effects. This provides a limited perspective * Correspondence: [email protected] 1 Iowa State University, Department of Kinesiology, Ames, IA 50011, USA Full list of author information is available at the end of the article
of the outcomes since it is possible for interventions to have differential effects within the sample population. An intervention, for example, might have benefits for inactive youth but not for others. It is therefore important to conduct follow-up evaluations of the data collected to examine possible moderating influences. Members of our team recently reported on the outcomes from a recess-based intervention, the Readyfor-Recess (R4R) project, a controlled trial evaluating environmental modifications to promote physical activity during recess [4]. Recess plays a critical role in children’s development and well being [5,6] and recess time has also been shown to make important contributions to children’s school and total daily activity [7-10].
© 2014 Saint-Maurice et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.
Saint-Maurice et al. BMC Public Health 2014, 14:103 http://www.biomedcentral.com/1471-2458/14/103
Ridgers et al., 2012, recently conducted a systematic review to examine activity levels during recess. The review emphasized the need to explore different activity promotion strategies and how these can impact the behaviors of important subgroups (e.g. boys vs. girls) [11]. Other recess-based studies have examined the benefits of adult supervision [12], and equipment availability during recess [13], but a unique aspect of R4R was that it evaluated the independent and combined effects of adult supervision and equipment on physical activity during recess. Schools in the R4R study (n = 12) were randomly assigned to one of four conditions: control group, staff supervision, provision of equipment, and the combination of staff supervision and equipment. The combination of staff supervision and equipment availability led to a significant increase of moderate-to-vigorous physical activity (MVPA) in boys (+14.1%) [4]. These findings suggest that recess interventions can be successful when promoting active behaviors among groups of children. As with most clinical trials [14], the success of R4R was examined using group-level comparisons with the rate of change across time being defined as a fixed effect. This approach assumes that the amount of change across individuals is the same. While this fixed effects approach is appropriate for the overall evaluation, further work is required so that the assumption of homogeneity across individuals can be examined and to avoid what is defined as the “ecological fallacy” [15,16]. The inter-individual variability in the context of a PA intervention can be examined by allowing the impact of the interventions to be defined as random in the regression model. A test of homogeneity is used to determine if the rate of change across time is consistent across individuals. If there is variability between individuals one can try to explain this variability and thereby better understand individual determinants of PA. This approach can be undertaken using latent growth curve analysis, which is similar to the classic fixed effects model but with the advantage that researchers can specify predictors in the model as random factors [16]. This approach makes it possible to systematically evaluate factors influencing the consistency of change across individuals. The purpose of this study was to examine if baseline levels of moderate-to-vigorous physical activity (MVPA) moderated the effects from the various R4R treatments.
Methods Intervention design
The R4R (August 2009-September 2010) project evaluated the independent and combined effects of staff training and equipment on children’s participation in MVPA during daily recess. Twelve schools were randomly assigned to one of four conditions: Staff Training (ST), Equipment (EQ), Staff Training + Equipment (STEQ) or
Page 2 of 8
a control condition. We have published a main outcome paper that provides additional details about the R4R intervention [4]. Participants
Elementary children (3rd-6th grades, 8–11 years of age) from 12 participating schools were recruited to participate in the evaluation of the R4R intervention. Demographic information obtained from the school included: age, gender, birth date, race, and free and reduced lunch status. The Ready for Recess intervention study was approved by the Institutional Review Board at University of Nebraska at Omaha. Children that returned signed consent forms were enrolled in the study. Anthropometric measures
Height and weight data on participants were obtained by the research team using standard field based techniques. However, we had to minimize the disruption of the school schedule at some of the schools and therefore were not able collect this information on the entire sample involved in the study. Physical activity - accelerometer measure
Physical Activity was assessed with the GT1M Actigraph Accelerometers (Actigraph, Pensacola, FL). Students were asked to wear the activity monitors during the school day (approximately 8:00 AM – 3:00 PM), for 4–5 consecutive days. Members of the research team put the monitors on the children each morning and took them off at the end of the school day to ensure compliance. Teachers provided logs of student attendance and a daily schedule to record time periods for recess, lunch, and physical education. The data from the Actigraphs were recorded in 5 second epochs and processed using Freedson et al. (2005) age-specific cutpoints (METs = 2.757 + (0.0015 · counts · min-1) – (0.08957 · age [yr]) – (0.000038 · counts · min1 · age [yr]) [17]. Each epoch during recess was classified as light (1.5 - 3.9 METs), moderate (MPA; 4.0 - 6.0 METs) or vigorous (VPA; > 6.0 METs) and the data were then aggregated to determine time spent in LPA, MPA, VPA and MVPA (MPA + VPA). Participants were included in the final analyses if they were at school (and had data) on at least 3 of the 5 days. Since the activity monitors were only used during school time it was assumed that students wore the monitors during all time unless indicated otherwise by school teachers. Activity outcomes were computed as an average of the valid data extracted from recess periods. The major computed outcome measure was the percent time allocated to MVPA in order to standardize for different lengths of recess.
Saint-Maurice et al. BMC Public Health 2014, 14:103 http://www.biomedcentral.com/1471-2458/14/103
Data analysis
Latent Growth Curve (LGC) models were used to account for the longitudinal nature of the data (repeatedmeasures) and potential individual differences (nested within schools) when exposed to the different treatments. The variable Trial (coded as 0 = trial 1, 1 = trial 2) was defined as a level-1 predictor (growth factor), characteristics of the children as level-2 predictors, and schools as level-3 predictors. Post-intervention average percent time in MVPA during recess (activity obtained from valid recess segments) was defined as the main outcome of interest (dependent variable). Level-2 covariates included grade (coded with 0 = 3rd grade, 1 = 4th grade, 2 = 5th grade, and 3 = 6th grade), gender (coded 0 = male and 1 = female), average recess duration (in minutes), and baseline (trial 1) percent time in MVPA during recess. The three treatments were used as level-3 predictors and coded using dummy coding (variable Staff coded 0 = No Staff, 1 = Staff; variable Equipment coded 0 = No equipment, 1 = Equipment; and variable Staff + Equipment coded 0 = No Staff + Equipment, and 1 = Staff + Equipment). Other demographic characteristics of the children such as BMI, race, or social economic status were not included since this information was not available for the entire sample. Only individuals with complete activity data at each assessment (Trial) were included in the analysis. To examine the direction, magnitude and consistency of the moderation of baseline MVPA on each intervention, least-square means comparing each treatment to the control group were computed (using the final LGC model) at three fixed values based on baseline MVPA scores: Mean - 1 standard deviation, Mean, and Mean + 1 standard deviation (see magnitude and direction of moderation effect). These three categories were operationalized as Low Active, Moderate Active, and High Active, respectively. These comparisons were performed to examine the overall trend over time (across treatments and control group) and also separately for each activity group. A total of 12 least-square means tests with their respective mean difference (Mean diff ), standard errors and p-values are reported. Results from the nested models (test of model fit for level-1 and level-2 predictors) and moderation effects were computed with 95% confidence intervals (p < .05). To account for the increased type I error associated with multiple statistical tests when comparing subgroups of activity, statistical analysis associated with least-square means were performed with a 99% confidence interval and therefore defined as significant if p-value < .01. In addition, mean differences for these analyses were considered borderline significant if p-value was between .01 and .05. Statistical procedures as described above were performed using SAS v9.2 (Cary, North Carolina).
Page 3 of 8
Results From the total sample of 667 participants, there were 393 participants (181 boys and 212 girls) who had three days of activity monitor data on both trials. The sample size differs from the main outcome results previously reported by Huberty et al., 2013, since it was important to have robust estimates of individual variability patterns across time. The distribution of participants across gender, grade and intervention is provided in Table 1. Inter-individual variability and test of covariates
Preliminary analyses were conducted to test assumptions needed for the use of LGC. We examined the variability in the MVPA scores across time to ensure that there was sufficient variability in the data. The intraclass correlation for individuals was equal to .08, while for schools it was equal to .07 and the covariance parameters indicated that intercept scores for time spent in MVPA were significantly different between individuals (see Figure 1) and schools (see Figure 2) (p < .05). Additional testing with level-1 and level-2 predictors revealed significant improvements in the fit of the model (indicated by a significant (p < .05) change in the (LLR) value). The results supported the clustered nature of the data and differences on baseline activity levels between individuals and schools. These findings support the planned model used to examine the moderator effect of baseline activity on change in MVPA levels associated with each condition. The latent growth curve model
Interpretation of the latent growth curve model b-weights (Table 2) revealed that change in MVPA from trial 1 to trial 2 in the control group (holding all the other predictors in the model constant) could be explained by initial MVPA scores. Per each unit increase in baseline MVPA score, there was a reduction of 0.76% on MVPA scores change from trial 1 to trial 2 (e.g., children in the control group with lower baseline MVPA scores had greater positive changes in MVPA from trial 1 to trial 2 while the opposite was true for children with higher baseline MVPA scores). Similar trial X baseline MVPA interactions with each condition revealed that this trend was attenuated by Table 1 Distribution (N) of gender and grade per intervention group Gender Boys
Girls
Grade 3rd
4th
5th
6th
Total
Control
47
45
20
27
24
21
92
ST
39
69
32
19
34
23
108
EQ
59
51
23
27
38
22
110
STEQ
36
47
8
22
23
30
83
Total
181
212
83
95
119
96
393
ST = Staff treatment; EQ = Equipment treatment; STEQ = Staff + Equipment treatment.
Saint-Maurice et al. BMC Public Health 2014, 14:103 http://www.biomedcentral.com/1471-2458/14/103
Page 4 of 8
Figure 1 Percent time spent in MVPA at recess for a (randomly selected) subgroup of 25 individuals. Each line represents an individual. This figure illustrates a high degree of variability in individual’s baseline scores (intercept) and change over time (slopes). This justifies the need for a random intercept term at level-2 (individual).
ST (β = 0.29 ± 0.12, p = .02) but not EQ (p = .35) or STEQ (p = .74). In other words, youth in the Staff condition that were less active at baseline had greater MVPA differences (from their respective control group) in MVPA at trial 2 when compared to their more active peers. Per each unit increase in percent MVPA at baseline, there was an attenuation of the ST effect by 0.47% in MVPA levels at trial 2. Magnitude and direction of the moderation effect
The overall mean score (across treatments and control group) for percent of time spent in MVPA at baseline was 27.5%. Predicted values were estimated by fixing baseline activity values in the final model as 27.5%, 12.5% (−1SD), and 42.4% (+1SD). These predicted activity scores represented Moderate, Low, and High Active groups, respectively. There were significant changes over time for the three groups but the patterns were very
different. The Low Active group had a significant increase in MVPA from trial 1 to trial 2 (Mean diff = 6.3 ± 0.9%; t (11) = 6.7; p < .001), while the Moderate (Mean diff = −4.5 ± 0.7%; t (11) = −6.8; p < .001) and High active group (Mean diff = −15.3 ± 0.9%; t (11) = −16.2; p < .001) decreased their MVPA levels (p < .001). Further evaluation, however, shows that the response varies depending on the treatment group. When stratified by treatment group, the mean baseline MVPA values (Medium) for the ST, EQ and STEQ treatments were 27.7%, 26.1%, and 27.1%, respectively. The corresponding values for Low Active (−1SD) were equal to, 12.2%, 11.0%, and 13.3% while values for High Active (+1SD) were 43.2%, 41.3%, and 40.8%. The values for Low Active were all similar at baseline (Trial 1) but the levels of MVPA at trial 2 in the ST schools were significantly lower than the respective Low Active control group
Figure 2 Percent time spent in MVPA at recess for the 12 schools. Each line represents a school. Similarly to Figure 1, there was a significant difference in baseline scores for school clusters, and therefore, justifying the need for random intercepts at this level.
Saint-Maurice et al. BMC Public Health 2014, 14:103 http://www.biomedcentral.com/1471-2458/14/103
Page 5 of 8
Table 2 Regression coefficients for the final latent growth curve model Intercept
Estimate
SE
−6.73
5.62 8
df
t-value p-value −1.20
0.27
b-weights Trial
20.35
3.00 759 6.78
